Skip to content

design: collegial coding agents (week-scale studio) - #68

Closed
headlong42 wants to merge 1 commit into
laude-institute:mainfrom
headlong42:audel/design-collegial-coding-agents
Closed

design: collegial coding agents (week-scale studio)#68
headlong42 wants to merge 1 commit into
laude-institute:mainfrom
headlong42:audel/design-collegial-coding-agents

Conversation

@headlong42

Copy link
Copy Markdown

Summary

Week-scale experiment: four Headlong identities on Hermes / Codex / Claude Code / Prime Agent, sharing one studio (same rails, same artifacts) and rotating a collegial propose → critique → integrate → reflect loop. Not a GAN, not an orchestrator splitting tickets.

Doc: design/collegial-coding-agents.md (plus a one-line index in design/README.md). No runtime changes.

Open questions before anyone spends money: which Prime, which shared problem, budget.

Test plan

  • Doc-only
  • Confirm Prime / problem / budget before phase 1

Proposal for four Headlong identities, each bound to a different
coding-agent backend (Hermes, Codex, Claude Code, Prime Agent),
sharing a generate-critique-integrate studio loop for one week.

Keeps Headlong as the infinite self-guided mind and the coding
agent as sandboxed hands. Collegial, not adversarial — GAN-shaped
loop without a discriminator to fool.
@bobbyhwilder

Copy link
Copy Markdown
Contributor

Superseded by #74 — a substantial reframe (temperature-not-authority asymmetry, disagree-and-commit decision split, reflexivity/self-modification, one-pair-first with a fleet as follow-on). Closing this in favor of #74.

@nickjalbert

Copy link
Copy Markdown
Contributor

Per #68 (comment), closing this in favor of #74

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants